1 // SPDX-License-Identifier: GPL-2.0
5 * (C) Copyright 1996 Linus Torvalds
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
12 #include <linux/hugetlb.h>
13 #include <linux/shm.h>
14 #include <linux/ksm.h>
15 #include <linux/mman.h>
16 #include <linux/swap.h>
17 #include <linux/capability.h>
19 #include <linux/swapops.h>
20 #include <linux/highmem.h>
21 #include <linux/security.h>
22 #include <linux/syscalls.h>
23 #include <linux/mmu_notifier.h>
24 #include <linux/uaccess.h>
25 #include <linux/mm-arch-hooks.h>
26 #include <linux/userfaultfd_k.h>
28 #include <asm/cacheflush.h>
29 #include <asm/tlbflush.h>
33 static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
40 pgd = pgd_offset(mm, addr);
41 if (pgd_none_or_clear_bad(pgd))
44 p4d = p4d_offset(pgd, addr);
45 if (p4d_none_or_clear_bad(p4d))
48 pud = pud_offset(p4d, addr);
49 if (pud_none_or_clear_bad(pud))
52 pmd = pmd_offset(pud, addr);
59 static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
67 pgd = pgd_offset(mm, addr);
68 p4d = p4d_alloc(mm, pgd, addr);
71 pud = pud_alloc(mm, p4d, addr);
75 pmd = pmd_alloc(mm, pud, addr);
79 VM_BUG_ON(pmd_trans_huge(*pmd));
84 static void take_rmap_locks(struct vm_area_struct *vma)
87 i_mmap_lock_write(vma->vm_file->f_mapping);
89 anon_vma_lock_write(vma->anon_vma);
92 static void drop_rmap_locks(struct vm_area_struct *vma)
95 anon_vma_unlock_write(vma->anon_vma);
97 i_mmap_unlock_write(vma->vm_file->f_mapping);
100 static pte_t move_soft_dirty_pte(pte_t pte)
103 * Set soft dirty bit so we can notice
104 * in userspace the ptes were moved.
106 #ifdef CONFIG_MEM_SOFT_DIRTY
107 if (pte_present(pte))
108 pte = pte_mksoft_dirty(pte);
109 else if (is_swap_pte(pte))
110 pte = pte_swp_mksoft_dirty(pte);
115 static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
116 unsigned long old_addr, unsigned long old_end,
117 struct vm_area_struct *new_vma, pmd_t *new_pmd,
118 unsigned long new_addr, bool need_rmap_locks, bool *need_flush)
120 struct mm_struct *mm = vma->vm_mm;
121 pte_t *old_pte, *new_pte, pte;
122 spinlock_t *old_ptl, *new_ptl;
123 bool force_flush = false;
124 unsigned long len = old_end - old_addr;
127 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
128 * locks to ensure that rmap will always observe either the old or the
129 * new ptes. This is the easiest way to avoid races with
130 * truncate_pagecache(), page migration, etc...
132 * When need_rmap_locks is false, we use other ways to avoid
135 * - During exec() shift_arg_pages(), we use a specially tagged vma
136 * which rmap call sites look for using is_vma_temporary_stack().
138 * - During mremap(), new_vma is often known to be placed after vma
139 * in rmap traversal order. This ensures rmap will always observe
140 * either the old pte, or the new pte, or both (the page table locks
141 * serialize access to individual ptes, but only rmap traversal
142 * order guarantees that we won't miss both the old and new ptes).
145 take_rmap_locks(vma);
148 * We don't have to worry about the ordering of src and dst
149 * pte locks because exclusive mmap_sem prevents deadlock.
151 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
152 new_pte = pte_offset_map(new_pmd, new_addr);
153 new_ptl = pte_lockptr(mm, new_pmd);
154 if (new_ptl != old_ptl)
155 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
156 flush_tlb_batched_pending(vma->vm_mm);
157 arch_enter_lazy_mmu_mode();
159 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
160 new_pte++, new_addr += PAGE_SIZE) {
161 if (pte_none(*old_pte))
164 pte = ptep_get_and_clear(mm, old_addr, old_pte);
166 * If we are remapping a dirty PTE, make sure
167 * to flush TLB before we drop the PTL for the
168 * old PTE or we may race with page_mkclean().
170 * This check has to be done after we removed the
171 * old PTE from page tables or another thread may
172 * dirty it after the check and before the removal.
174 if (pte_present(pte) && pte_dirty(pte))
176 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
177 pte = move_soft_dirty_pte(pte);
178 set_pte_at(mm, new_addr, new_pte, pte);
181 arch_leave_lazy_mmu_mode();
182 if (new_ptl != old_ptl)
183 spin_unlock(new_ptl);
184 pte_unmap(new_pte - 1);
186 flush_tlb_range(vma, old_end - len, old_end);
189 pte_unmap_unlock(old_pte - 1, old_ptl);
191 drop_rmap_locks(vma);
194 #define LATENCY_LIMIT (64 * PAGE_SIZE)
196 unsigned long move_page_tables(struct vm_area_struct *vma,
197 unsigned long old_addr, struct vm_area_struct *new_vma,
198 unsigned long new_addr, unsigned long len,
199 bool need_rmap_locks)
201 unsigned long extent, next, old_end;
202 pmd_t *old_pmd, *new_pmd;
203 bool need_flush = false;
204 unsigned long mmun_start; /* For mmu_notifiers */
205 unsigned long mmun_end; /* For mmu_notifiers */
207 old_end = old_addr + len;
208 flush_cache_range(vma, old_addr, old_end);
210 mmun_start = old_addr;
212 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
214 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
216 next = (old_addr + PMD_SIZE) & PMD_MASK;
217 /* even if next overflowed, extent below will be ok */
218 extent = next - old_addr;
219 if (extent > old_end - old_addr)
220 extent = old_end - old_addr;
221 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
224 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
227 if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {
228 if (extent == HPAGE_PMD_SIZE) {
230 /* See comment in move_ptes() */
232 take_rmap_locks(vma);
233 moved = move_huge_pmd(vma, old_addr, new_addr,
234 old_end, old_pmd, new_pmd,
237 drop_rmap_locks(vma);
241 split_huge_pmd(vma, old_pmd, old_addr);
242 if (pmd_trans_unstable(old_pmd))
245 if (pte_alloc(new_vma->vm_mm, new_pmd, new_addr))
247 next = (new_addr + PMD_SIZE) & PMD_MASK;
248 if (extent > next - new_addr)
249 extent = next - new_addr;
250 if (extent > LATENCY_LIMIT)
251 extent = LATENCY_LIMIT;
252 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
253 new_pmd, new_addr, need_rmap_locks, &need_flush);
256 flush_tlb_range(vma, old_end-len, old_addr);
258 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
260 return len + old_addr - old_end; /* how much done */
263 static unsigned long move_vma(struct vm_area_struct *vma,
264 unsigned long old_addr, unsigned long old_len,
265 unsigned long new_len, unsigned long new_addr,
266 bool *locked, struct vm_userfaultfd_ctx *uf,
267 struct list_head *uf_unmap)
269 struct mm_struct *mm = vma->vm_mm;
270 struct vm_area_struct *new_vma;
271 unsigned long vm_flags = vma->vm_flags;
272 unsigned long new_pgoff;
273 unsigned long moved_len;
274 unsigned long excess = 0;
275 unsigned long hiwater_vm;
278 bool need_rmap_locks;
281 * We'd prefer to avoid failure later on in do_munmap:
282 * which may split one vma into three before unmapping.
284 if (mm->map_count >= sysctl_max_map_count - 3)
288 * Advise KSM to break any KSM pages in the area to be moved:
289 * it would be confusing if they were to turn up at the new
290 * location, where they happen to coincide with different KSM
291 * pages recently unmapped. But leave vma->vm_flags as it was,
292 * so KSM can come around to merge on vma and new_vma afterwards.
294 err = ksm_madvise(vma, old_addr, old_addr + old_len,
295 MADV_UNMERGEABLE, &vm_flags);
299 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
300 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
305 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
307 if (moved_len < old_len) {
309 } else if (vma->vm_ops && vma->vm_ops->mremap) {
310 err = vma->vm_ops->mremap(new_vma);
315 * On error, move entries back from new area to old,
316 * which will succeed since page tables still there,
317 * and then proceed to unmap new area instead of old.
319 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
326 mremap_userfaultfd_prep(new_vma, uf);
327 arch_remap(mm, old_addr, old_addr + old_len,
328 new_addr, new_addr + new_len);
331 /* Conceal VM_ACCOUNT so old reservation is not undone */
332 if (vm_flags & VM_ACCOUNT) {
333 vma->vm_flags &= ~VM_ACCOUNT;
334 excess = vma->vm_end - vma->vm_start - old_len;
335 if (old_addr > vma->vm_start &&
336 old_addr + old_len < vma->vm_end)
341 * If we failed to move page tables we still do total_vm increment
342 * since do_munmap() will decrement it by old_len == new_len.
344 * Since total_vm is about to be raised artificially high for a
345 * moment, we need to restore high watermark afterwards: if stats
346 * are taken meanwhile, total_vm and hiwater_vm appear too high.
347 * If this were a serious issue, we'd add a flag to do_munmap().
349 hiwater_vm = mm->hiwater_vm;
350 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
352 /* Tell pfnmap has moved from this vma */
353 if (unlikely(vma->vm_flags & VM_PFNMAP))
354 untrack_pfn_moved(vma);
356 if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) {
357 /* OOM: unable to split vma, just get accounts right */
358 vm_unacct_memory(excess >> PAGE_SHIFT);
361 mm->hiwater_vm = hiwater_vm;
363 /* Restore VM_ACCOUNT if one or two pieces of vma left */
365 vma->vm_flags |= VM_ACCOUNT;
367 vma->vm_next->vm_flags |= VM_ACCOUNT;
370 if (vm_flags & VM_LOCKED) {
371 mm->locked_vm += new_len >> PAGE_SHIFT;
378 static struct vm_area_struct *vma_to_resize(unsigned long addr,
379 unsigned long old_len, unsigned long new_len, unsigned long *p)
381 struct mm_struct *mm = current->mm;
382 struct vm_area_struct *vma = find_vma(mm, addr);
385 if (!vma || vma->vm_start > addr)
386 return ERR_PTR(-EFAULT);
389 * !old_len is a special case where an attempt is made to 'duplicate'
390 * a mapping. This makes no sense for private mappings as it will
391 * instead create a fresh/new mapping unrelated to the original. This
392 * is contrary to the basic idea of mremap which creates new mappings
393 * based on the original. There are no known use cases for this
394 * behavior. As a result, fail such attempts.
396 if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) {
397 pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n", current->comm, current->pid);
398 return ERR_PTR(-EINVAL);
401 if (is_vm_hugetlb_page(vma))
402 return ERR_PTR(-EINVAL);
404 /* We can't remap across vm area boundaries */
405 if (old_len > vma->vm_end - addr)
406 return ERR_PTR(-EFAULT);
408 if (new_len == old_len)
411 /* Need to be careful about a growing mapping */
412 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
413 pgoff += vma->vm_pgoff;
414 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
415 return ERR_PTR(-EINVAL);
417 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
418 return ERR_PTR(-EFAULT);
420 if (vma->vm_flags & VM_LOCKED) {
421 unsigned long locked, lock_limit;
422 locked = mm->locked_vm << PAGE_SHIFT;
423 lock_limit = rlimit(RLIMIT_MEMLOCK);
424 locked += new_len - old_len;
425 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
426 return ERR_PTR(-EAGAIN);
429 if (!may_expand_vm(mm, vma->vm_flags,
430 (new_len - old_len) >> PAGE_SHIFT))
431 return ERR_PTR(-ENOMEM);
433 if (vma->vm_flags & VM_ACCOUNT) {
434 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
435 if (security_vm_enough_memory_mm(mm, charged))
436 return ERR_PTR(-ENOMEM);
443 static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
444 unsigned long new_addr, unsigned long new_len, bool *locked,
445 struct vm_userfaultfd_ctx *uf,
446 struct list_head *uf_unmap_early,
447 struct list_head *uf_unmap)
449 struct mm_struct *mm = current->mm;
450 struct vm_area_struct *vma;
451 unsigned long ret = -EINVAL;
452 unsigned long charged = 0;
453 unsigned long map_flags;
455 if (offset_in_page(new_addr))
458 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
461 /* Ensure the old/new locations do not overlap */
462 if (addr + old_len > new_addr && new_addr + new_len > addr)
465 ret = do_munmap(mm, new_addr, new_len, uf_unmap_early);
469 if (old_len >= new_len) {
470 ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap);
471 if (ret && old_len != new_len)
476 vma = vma_to_resize(addr, old_len, new_len, &charged);
482 map_flags = MAP_FIXED;
483 if (vma->vm_flags & VM_MAYSHARE)
484 map_flags |= MAP_SHARED;
486 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
487 ((addr - vma->vm_start) >> PAGE_SHIFT),
489 if (offset_in_page(ret))
492 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, uf,
494 if (!(offset_in_page(ret)))
497 vm_unacct_memory(charged);
503 static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
505 unsigned long end = vma->vm_end + delta;
506 if (end < vma->vm_end) /* overflow */
508 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
510 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
511 0, MAP_FIXED) & ~PAGE_MASK)
517 * Expand (or shrink) an existing mapping, potentially moving it at the
518 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
520 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
521 * This option implies MREMAP_MAYMOVE.
523 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
524 unsigned long, new_len, unsigned long, flags,
525 unsigned long, new_addr)
527 struct mm_struct *mm = current->mm;
528 struct vm_area_struct *vma;
529 unsigned long ret = -EINVAL;
530 unsigned long charged = 0;
532 struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX;
533 LIST_HEAD(uf_unmap_early);
536 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
539 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
542 if (offset_in_page(addr))
545 old_len = PAGE_ALIGN(old_len);
546 new_len = PAGE_ALIGN(new_len);
549 * We allow a zero old-len as a special case
550 * for DOS-emu "duplicate shm area" thing. But
551 * a zero new-len is nonsensical.
556 if (down_write_killable(¤t->mm->mmap_sem))
559 if (flags & MREMAP_FIXED) {
560 ret = mremap_to(addr, old_len, new_addr, new_len,
561 &locked, &uf, &uf_unmap_early, &uf_unmap);
566 * Always allow a shrinking remap: that just unmaps
567 * the unnecessary pages..
568 * do_munmap does all the needed commit accounting
570 if (old_len >= new_len) {
571 ret = do_munmap(mm, addr+new_len, old_len - new_len, &uf_unmap);
572 if (ret && old_len != new_len)
579 * Ok, we need to grow..
581 vma = vma_to_resize(addr, old_len, new_len, &charged);
587 /* old_len exactly to the end of the area..
589 if (old_len == vma->vm_end - addr) {
590 /* can we just expand the current mapping? */
591 if (vma_expandable(vma, new_len - old_len)) {
592 int pages = (new_len - old_len) >> PAGE_SHIFT;
594 if (vma_adjust(vma, vma->vm_start, addr + new_len,
595 vma->vm_pgoff, NULL)) {
600 vm_stat_account(mm, vma->vm_flags, pages);
601 if (vma->vm_flags & VM_LOCKED) {
602 mm->locked_vm += pages;
612 * We weren't able to just expand or shrink the area,
613 * we need to create a new one and move it..
616 if (flags & MREMAP_MAYMOVE) {
617 unsigned long map_flags = 0;
618 if (vma->vm_flags & VM_MAYSHARE)
619 map_flags |= MAP_SHARED;
621 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
623 ((addr - vma->vm_start) >> PAGE_SHIFT),
625 if (offset_in_page(new_addr)) {
630 ret = move_vma(vma, addr, old_len, new_len, new_addr,
631 &locked, &uf, &uf_unmap);
634 if (offset_in_page(ret)) {
635 vm_unacct_memory(charged);
638 up_write(¤t->mm->mmap_sem);
639 if (locked && new_len > old_len)
640 mm_populate(new_addr + old_len, new_len - old_len);
641 userfaultfd_unmap_complete(mm, &uf_unmap_early);
642 mremap_userfaultfd_complete(&uf, addr, new_addr, old_len);
643 userfaultfd_unmap_complete(mm, &uf_unmap);